Deploying Claude Code on a VPS has become a highly practical trend among developers in 2026. Whether you’re tweaking a few lines of code on your tablet during your commute, letting an AI agent continue tasks in the cloud after you’ve clocked off, or shifting your entire development environment from a local laptop to a server that stays online 24/7—all these scenarios lead to the same question: what kind of CPU and RAM specifications does your VPS actually need?
The answers you find online are all over the place. Some claim it can run on a single core with 1GB of RAM, while others insist you shouldn’t even bother with less than 16GB. Both perspectives are technically correct; the distinction lies in the difference between simply "getting it to run" and "ensuring stable, reliable operation."
First, let’s look at what Claude Code actually consumes.
Claude Code is, at its core, a command-line tool. Its primary functions include receiving instructions, calling Anthropic’s cloud-based APIs, and executing shell commands or file operations locally. It does not handle model inference itself—meaning it has no requirement for a GPU—which makes it fundamentally different from running a large language model (LLM) locally.
However, its resource consumption pattern is unique. According to Anthropic’s official documentation, Claude Code requires at least 4GB of RAM and an x64 or ARM64 processor. While the installation process only needs about 512MB of free memory, actual usage scales upward as task complexity increases.
When idle, a typical agent consumes roughly 300–600MB of RAM. During complex tasks—such as multi-process deployments or operations involving calls to MCP (Model Context Protocol) tools—memory usage can peak above 1.5GB. If you are simultaneously running Docker containers, Node.js services, or databases, these memory demands will quickly compound.
More importantly, Claude Code is known to have issues with memory growth. Users on GitHub have reported instances where a single Claude Code session—during an 18-minute, multi-step coding task—saw its RSS (resident set size) memory climb to 29.9GB, eventually triggering the system’s OOM (Out-of-Memory) Killer to terminate the process. While this is an extreme case not every user will encounter, it highlights a crucial reality: the upper limit of Claude Code’s memory consumption is significantly higher than its "stated requirements."
CPU: Single-core performance matters more than core count.
Claude Code represents a classic I/O-intensive workload, rather than a compute-intensive one. Most of the time is spent waiting—waiting for API responses, the results of shell command executions, or the completion of file read/write operations—rather than on intensive CPU computation.
This means that a dual-core, high-frequency CPU often provides a better experience than an eight-core, low-frequency one. Adding more cores won't lead to higher utilization, as most of Claude Code’s workflows are sequential. What truly impacts response latency is the single-core clock speed and the processor's architectural generation.
Take Jtti’s Hong Kong "lightweight" cloud instances as an example: their 1-core, 1GB RAM configuration uses an Intel Xeon E5-2680 v3 (2.50 GHz), which delivers stable single-core performance for lightweight tasks. This level of CPU power is sufficient for individual developers, single-session usage, basic code completion, and file operations.
However, if your workflow involves running multiple simultaneous Claude Code sessions, or if you are using Claude Code alongside Docker containers or local development servers, a dual-core setup will offer more headroom. Jtti’s 2-core/2GB or 2-core/4GB plans provide the appropriate level of computing power for such scenarios.
In short: the number of CPU cores is not the bottleneck; single-core frequency and the capacity to handle concurrency are what matter. One core is enough, but two cores are a safer, more reliable choice.
Memory: 4 GB is the bare minimum; 8 GB is the "comfort zone."
Memory is the resource that requires the most careful consideration when deploying Claude Code. Anthropic specifies 4 GB as the absolute minimum—which simply means it will "run," not that it will run "well."
1 GB of RAM: Suitable only for simple, scheduled scripts or very lightweight testing. The combined overhead of Claude Code and the operating system may push usage to the limit; as soon as a slightly more complex task is initiated, the OOM (Out-of-Memory) Killer will likely terminate the process.
2 GB of RAM: Sufficient to launch and run Claude Code, though it will require the use of swap space. If you are working with multiple GitHub projects, it is recommended to configure at least 2 GB of swap space to handle sudden memory spikes. While a 2 GB machine can handle light tasks, the reliance on swap will become a performance bottleneck if you introduce Docker or multi-process operations. 4 GB RAM: This is the minimum requirement officially stated by Anthropic and the baseline for "functional" usage. Testing by Cloudzy indicates that while Code Server alone can run on 1 GB, Claude Code—with its full technology stack—raises the minimum requirement to 4 GB; an 8 GB setup is recommended for a better experience.
8 GB RAM: The recommended configuration. This capacity allows Claude Code to operate alongside multi-container environments (such as those managed by Docker Compose) and provides a sufficient buffer for memory spikes. If you intend to use Claude Code as a core tool in your daily development workflow, 8 GB is the more sensible choice.
16 GB RAM and above: Suitable for scenarios involving multiple users, the simultaneous execution of several AI agents, or the need to run additional local services (such as databases, caches, or message queues) alongside Claude Code.
An often-overlooked detail: Setting a memory "fuse" for Claude Code
Given that Claude Code’s memory consumption can occasionally spike, it is wise to prevent it from consuming unlimited resources. On Linux, the most effective approach is to enforce a hard memory limit using `systemd` cgroups:
systemd-run --user --scope -p MemoryMax=4G -p MemorySwapMax=0 -- claudeThis command caps Claude Code’s memory usage at 4 GB. If the process exceeds this limit, the OOM (Out-Of-Memory) Killer will terminate it, preventing it from crashing the entire server. This is particularly critical for a VPS with only 4 GB of RAM—it ensures that if an anomaly occurs, Claude Code "dies" on its own rather than taking the entire system down with it.
Recommended configurations for Jtti service nodes
Based on the resource analysis above, Jtti’s product lineup can be matched to various use cases:
Individual developers / Light usage (1 vCPU, 1 GB RAM): Jtti’s Hong Kong "Lightweight Cloud" (1 vCPU, 1 GB RAM, 200 Mbps dedicated bandwidth, 800 GB monthly data transfer) is a suitable option. The 200 Mbps dedicated bandwidth is more than sufficient for Claude Code to make API calls and fetch dependencies, and the data allowance covers typical daily development needs. Since 1 GB is the absolute minimum, it is advisable to configure swap space and use `cgroup` limits; this setup is best suited for tasks like running remote scripts or making occasional code edits. For daily development and running multiple concurrent containers (2 vCPU/2GB or 2 vCPU/4GB RAM): Jtti’s Hong Kong CN2 VPS offers a 2 vCPU/2GB RAM/2Mbps bandwidth plan for $118.91/year. The 2 vCPU/4GB RAM/5Mbps plan is well-suited for environments where Claude Code and Docker containers coexist. The CN2 GIA network—which provides direct connectivity across all three major Chinese carriers—ensures low latency for SSH connections and API calls from within China, preventing coding workflows from being interrupted by cross-border network instability.
For heavy workloads and multiple concurrent agents (4 vCPU/8GB RAM or higher): Jtti offers a US-based CN2 GIA plan (4 vCPU/8GB RAM/10Mbps, unmetered traffic) for $360.32/year, or a Hong Kong-based plan (4 vCPU/8GB RAM/5Mbps) for $345.41/year. With 8GB of RAM, you have ample headroom for Claude Code’s multi-session operations, complex MCP (Model Context Protocol) tool calls, and parallel Docker containers.
All plans come with a recurring lifetime discount, meaning the price you pay for your initial purchase remains the same for all subsequent renewals. For a development environment like Claude Code—which requires long-term, continuous uptime—predictable costs are a key factor in choosing the right infrastructure.
When running Claude Code on a Jtti VPS, the logic for selecting a configuration can be summarized as follows: evaluate the CPU based on single-core speed, and assess the RAM based on how much "headroom" remains above the minimum requirement.
A 1 vCPU/1GB RAM setup is functional but only suitable for lightweight tasks. A 2 vCPU/2GB RAM configuration serves as an entry-level baseline, though it will require swap space as a safety net. While 4GB of RAM is the minimum threshold recommended by Anthropic, 8GB is the "sweet spot" for long-term, comfortable use. Regardless of the configuration you choose, it is essential to set a `cgroup` memory limit for Claude Code; this acts as a safeguard, ensuring that an anomalous process doesn't crash the entire server.
Visit the official Jtti website to view the full range of configurations and test IPs for their Hong Kong and US cloud server nodes, and select the plan that best aligns with your specific Claude Code usage intensity.
